AMAZON 2023 · COMPLETED RESEARCH

Inspect the public-trace footprint.

Explore Gijs Overgoor and coauthors’ completed analysis of the Amazon Reviews 2023 archive, collected by McAuley Lab at UC San Diego. Every setting below opens saved research results.

Source: McAuley Lab’s Amazon Reviews 2023 dataset on Hugging Face ↗

Original research · recorded results

Choose a result

A reviewer joins the cohort by reaching this count within one 100-history session. Each count records support from a distinct opened reviewer neighborhood. Membership is then combined across sessions.

What stays fixed?

These completed runs opened up to 100 histories per session, with an expansion score cap of 5. That cap guides the next history to open; it is separate from the cohort entry count.

Bursts use a ±28-day window around a focal review. Recurrence requires at least two burst products.

Method and paper →

SELECTED RESEARCH SCOPE

All five-star reviews

January 1–September 14, 2023 · full-release archive audit

Cohort entry count 30Minimum burst 10 reviews

The complete procedure includes all five-star reviews in the audit period, whether or not the purchase was marked verified.

L0 · REVIEW FOOTPRINT1.4%

Cohort-authored reviews

225,267 reviews

Any five-star review by a cohort member.

L1 · REVIEW FOOTPRINT0.1%

Reviews in cohort bursts

9,702 reviews

Cohort-authored reviews with the selected minimum number of cohort reviews on the same product within 28 days before or after the review date.

L2 · REVIEW FOOTPRINT0.1%

Reviews by recurring burst authors

9,350 reviews

Reviews in cohort bursts whose author appears in qualifying bursts on at least two distinct products.

OVERALL DENOMINATOR16,035,782

Five-star reviews

Used for the overall, entry-count, and ordering results. Categories use their own denominators.

Jan 1–Sep 14, 2023

L0, L1, and L2 are nested footprints of public review structure. They are not additive and do not estimate confirmed fraud or an individual account’s fraud probability.

COHORT ENTRY COUNT

How the footprint changes.

7 completed settings

The review scope and minimum burst size stay as selected above. Each row shows a saved result for a different cohort entry count.

Review footprint at seven cohort entry counts, within the selected scope and burst size
Entry count	L0 share	L1 share	L2 share
2	79.9%	35.2%	20.6%
5	66.2%	29.7%	17.6%
10	43.4%	17.5%	11.2%
14	10.9%	2.5%	2.4%
18	7.0%	1.1%	1.0%
20	5.3%	0.7%	0.7%
30Selected	1.4%	0.1%	0.1%

L0: reviews authored by the cohort. L1: cohort reviews in bursts. L2: cohort-burst reviews by recurring authors. Bars use the same 0–100% scale.

ACROSS PRODUCT CATEGORIES

The footprint by category.

The completed category breakdown uses cohort entry count 10 and a minimum burst of 5. It is not available at the settings currently selected.

Show the completed category setting →

SEARCH ORDER

Variation across completed orderings.

The three completed ordering runs use cohort entry count 10 and a minimum burst of 5. No ordering range is shown for other settings.

Show the completed ordering setting →

RESEARCH AND DATA

Explore the findings and the research behind them.

Source archive: McAuley Lab’s Amazon Reviews 2023 on Hugging Face ↗. The release contains reviews from multiple years. The reported audit uses reviews dated January 1–September 14, 2023 across all 34 release categories, including Unknown.

Analysis: Gijs Overgoor and coauthors applied the paper’s method to that archive to calculate the findings shown here. The earlier 2019–2020 campaign dataset is a separate public resource:

Public research dataset ↗

These are the paper’s existing archive results. This site does not accept online contributions or provide account-level historical membership checks. Use the investigator to collect and export a case locally.